Skip to content

Datasets

PME-toolkit operates on datasets stored in external files (typically .mat) and referenced through configuration files.

Dataset organization

Dataset metadata is documented in:

  • databases/glider/
  • databases/airfoil/

These folders describe:

  • dataset structure
  • variable definitions
  • required files

Benchmark datasets

The main benchmark datasets are distributed via Zenodo:

Glider

Serani, A., & Palma, G. (2026).
Design-Space Dimensionality Reduction Benchmark Dataset – Bio-Inspired Underwater Glider (1.1) [Data set].
https://doi.org/10.5281/zenodo.18936594

Airfoil

Serani, A., & Quagliarella, D. (2026).
Design-Space Dimensionality Reduction Benchmark Dataset – RAE2822 Airfoil (1.0) [Data set].
https://doi.org/10.5281/zenodo.18958555

Dataset file

The dataset is specified in the configuration file:

"io": {
    "dbfile": "path/to/database.mat"
}

The file typically contains:

  • geometric data
  • design variables
  • optional physical quantities

Data structure

Datasets are organized sample-wise, where:

  • each column represents one design configuration
  • all data sources are aligned across samples

The dataset may include:

Geometry

  • discretized coordinates
  • structured according to geom definition

Variables

  • parametric design variables
  • dimension defined by vars.Mbase

Physics (optional)

  • distributed fields (e.g. pressure coefficients)
  • scalar quantities (e.g. drag, lift)

Consistency requirements

All dataset components must:

  • have the same number of samples
  • be aligned across geometry, variables, and physics
  • be free of missing values (after filtering)

Usage

To use a dataset:

  1. download the dataset from Zenodo
  2. extract it locally
  3. update the dbfile path in the JSON configuration
  4. run the benchmark

Notes

  • dataset structure must match the configuration file
  • geometry, variables, and physics must match geom, vars, and phys
  • incorrect alignment will lead to invalid results

Summary

Datasets provide the input to PME workflows and must be:

  • versioned (via Zenodo)
  • structured consistently
  • correctly referenced in configuration files